[f-nsp] CER bfd

[Youssef Bengelloun-Zahr]

Hello,

Software, unfortunalty. This is something I got recently from the Brocade
support when I opened a case :

"in 5.4 & 5.5 code , BFD packet processing is done entirely in SW (with no
HW assist). If the CPU is too busy, and a process is taking too much time
to perform certain tasks (such as programming many routes), BFD packet
processing can be delayed (in TX and RX directions), that then result in
BFD flap.


What has improved in 5.6

Three things have been done to alleviate this situation in 5.6.
1.   HW assist has been implemented in CER for BFD packets. This means that
BFD packets are transmitted by HW.
2.   CPU optimizations have been added in BGP code so as not to use CPU for
too long without yielding. This should help with flaps that were happening
during route updates.
3.   Other tasks have also been optimized to reduce their impact on all
time-sensitive protocols.
"

I know, it's annoying.

HTH.

BR.




2014-02-18 11:47 GMT+01:00 Darren O'Connor <darrenoc at outlook.com>:

> Can anyone confirm if the CER does BFD in hardware or software? The config
> guide states 20 sessions, but no mention on LP or CPU based hellos
>
> I've been seeing BFD flap on a cer when rancid logs and and gets config
>
> Thanks
>
> http://www.mellowd.co.uk/ccie
>
> _______________________________________________
> foundry-nsp mailing list
> foundry-nsp at puck.nether.net
> http://puck.nether.net/mailman/listinfo/foundry-nsp
>



-- 
Youssef BENGELLOUN-ZAHR
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://puck.nether.net/pipermail/foundry-nsp/attachments/20140218/2fb11c22/attachment.html>